머신러닝 기반 항공기 사고 기체 손상 심각도에 영향을 미치는 요인 분석

초록

This study examines aircraft damage severity by framing it as a machine learning–based multiclass classification problem. Using NTSB accident records, we constructed a multidimensional dataset that integrates aircraft, pilot, airport, and meteorological information. XGBoost was applied under class-imbalanced conditions, with recall as the primary performance metric and SHAP analysis used to enhance model interpretability. The results indicate that aircraft damage severity is driven by multiple interacting factors, particularly spatial airport-related variables. Among these factors, airport elevation and spatial proximity measures were more influential than traditional predictors such as aircraft age or general weather conditions. These findings underscore the need for a shift toward proactive aviation safety management based on aircraft damage–oriented risk assessment.

키워드